Early Cholangiocarcinoma Detection With Magnetic Resonance Imaging Versus Ultrasound in Primary Sclerosing Cholangitis

Abstract

BACKGROUND AND AIMS:

Early detection of perihilar cholangiocarcinoma (CCA) among patients with primary sclerosing cholangitis (PSC) is important to identify more people eligible for curative therapy. While many recommend CCA screening, there are divergent opinions and limited data regarding the use of ultrasound or magnetic resonance imaging (MRI) for early CCA detection, and it is unknown whether there is benefit in testing asymptomatic individuals. Our aims were to assess the diagnostic performances and prognostic implications of ultrasound and MRI-based CCA detection.

APPROACH AND RESULTS:

This is a multicenter review of 266 adults with PSC (CCA, n = 120) who underwent both an ultrasound and MRI within 3 months. Images were re-examined by radiologists who were blinded to the clinical information. Respectively, MRI had a higher area under the curve compared with ultrasound for CCA detection: 0.87 versus 0.70 for the entire cohort; 0.81 versus 0.59 for asymptomatic individuals; and 0.88 versus 0.71 for those listed for CCA transplant protocol. The absence of symptoms at CCA diagnosis was associated with improved 5-year outcomes including overall survival (82% vs. 46%, log-rank P < 0.01) and recurrence-free survival following liver transplant (89% vs. 65%, log-rank P = 0.04). Among those with asymptomatic CCA, MRI detection (compared with ultrasound) was associated with reduction in both mortality (hazard ratio, 0.10; 95% confidence interval, 0.01-0.96) and CCA progression after transplant listing (hazard ratio, 0.10; 95% confidence interval, 0.01-0.90). These benefits continued among patients who had annual monitoring and PSC for more than 1 year before CCA was diagnosed.

CONCLUSIONS:

MRI is superior to ultrasound for the detection of early-stage CCA in patients with PSC. Identification of CCA before the onset of symptoms with MRI is associated with improved outcomes.

Cholangiocarcinoma (CCA) is prevalent among those with primary sclerosing cholangitis (PSC), and early detection is important. Indeed, 10%-15% of patients with PSC may develop CCA, typically the perihilar subtype. However, early-stage perihilar CCA is challenging to diagnose through noninvasive imaging, because a discrete mass can be absent, and distinguishing benign from malignant strictures is difficult.^(1,2) CCA is the leading cause of death in this patient population.⁽³⁾ In part, this is due to the delayed detection of CCA until advanced incurable stages. Neoadjuvant chemotherapy and radiation followed by liver transplantation is an effective and well-studied treatment for early-stage perihilar CCA among those with PSC.⁽⁴⁾ This Medicare-approved approach has yielded a 5-year recurrence-free survival of 72% and is a standard indication for Model for End-Stage Liver Disease (MELD) exception points, which has been systematically studied across multiple centers.⁽⁴⁾ However, only persons with early-stage malignancy can benefit from this treatment. Consequently, early detection has the potential to increase the pool of patients eligible for curative therapy and thereby improve outcomes.

An optimal imaging test to detect CCA would be able to diagnose it in asymptomatic persons at an early stage with a high sensitivity. Such an approach can be used to triage patients toward confirmatory invasive testing with an endoscopic retrograde cholangiopancreatography (ERCP) with biliary brushings. To use a program for early detection, it is paramount to understand the diagnostic performances of two principle noninvasive imaging modalities: ultrasound and magnetic resonance imaging (MRI) with magnetic resonance cholangiopancreatography (MRCP). Indeed, the use of ultrasound is well established for hepatocellular carcinoma (HCC) surveillance among those with cirrhosis and is relatively inexpensive, whereas MRI with MRCP is noninvasive, avoids radiation, and is the preferred radiologic modality to image the biliary tree. In contrast, computed tomography (CT) is associated with radiation exposure and provides limited imaging of the biliary tree. To date, there has been one study that directly compared ultrasound to MRI for CCA detection among those with PSC. However, this study had a small cohort of cancer patients (n = 23) that was enriched with symptomatic individuals with advanced disease.⁽⁵⁾ Consequently, our knowledge of the diagnostic performances of ultrasound and MRI to detect early-stage curable CCA is limited.

It is important to demonstrate whether early CCA detection translates into improved outcomes and whether there is benefit to testing asymptomatic patients. Indeed, there is a lack of evidence to support the practice of CCA surveillance among asymptomatic individuals, and it is unclear whether imaging studies should be used only when symptoms of a progressive biliary obstruction develop. A prior study examined the routine use of ultrasound, MRI, or CT scans to detect a variety of hepatobiliary cancers in patients with PSC (n = 79).⁽⁶⁾ Individuals who sought routine care and had regular imaging had improved outcomes compared with those who did not. This study has several features to highlight. First, most patients had differing radiologic studies performed that precluded a direct comparison of one imaging test versus another. Second, this study pooled a variety of malignancies (intrahepatic CCA, gallbladder cancer, HCC) that are easier to diagnose and have varied approaches to treatment and outcomes, with the more prevalent but challenging to detect perihilar CCA. Finally, it is important to note that among the 14 subjects with perihilar CCA in the “surveillance” group, many already had symptoms of a progressive biliary obstruction that would have otherwise justified diagnostic imaging. Consequently, both the prognostic implications of asymptomatic perihilar CCA recognition and whether detection with one imaging modality versus another is associated with improved clinical outcomes remain unknown.

The lack of evidence to support the practice of routine imaging for early perihilar CCA detection has led to divergent opinions. For example, the American Association for the Study of Liver Diseases does not recommend for or against CCA surveillance.⁽⁷⁾ Alternatively, some suggest ultrasound can be used for CCA surveillance,^(8,9) while others advocate the use of MRI.⁽²⁾ Consequently, we sought to directly compare the diagnostic performances of ultrasound and MRI to detect early-stage perihilar CCA among those with PSC, determine whether it would be advantageous to diagnose CCA in asymptomatic patients (vs. symptomatic), and assess whether one imaging modality compared with another offers a prognostic benefit.

Materials and Methods

PATIENTS AND INDICATIONS FOR IMAGING

This study was approved by the institutional review board at Mayo Clinic, Rochester, MN, and conforms to the ethical guidelines of the 1975 Declaration of Helsinki. The IRB deemed this protocol minimal risk and waived need for informed consent.

Adults with large-duct PSC who underwent both an ultrasound and MRI of the abdomen within 3 months of each other were considered for inclusion, and the medical records of these individuals were reviewed in detail.⁽⁷⁾ During this study interval, serial CT scans were not routinely performed at our centers for CCA surveillance, and this imaging modality was not considered. Subjects were seen between the years 2000 and 2018 at the Mayo Clinic in either Rochester, MN, Jacksonville, FL, or Scottsdale, AZ. Patients with CCA were required to have their imaging studies within 3 months of the cancer diagnosis, and individuals without cancer were required to have a minimum follow-up of 2 years after imaging to prevent the possibility of an occult CCA. To minimize heterogeneity and bias, individuals with a hepatobiliary malignancy aside from perihilar CCA (distal or intrahepatic CCA, gallbladder cancer, or HCC) were excluded. Subjects were excluded if there was a concern for underlying CCA based on indeterminate biliary brush cytology and serial or multifocal polysomy on fluorescence in situ hybridization (FISH) but did not meet the subsequent definition of CCA. Patients with imaging studies deemed to be suboptimal by the blinded radiology team were also excluded.

Individuals with PSC are seen in our general hepatobiliary clinics or transplant centers. Patients with PSC underwent both an ultrasound and MRI/MRCP as part of their initial transplant clinic consultation. Performing these imaging studies concomitantly was a prospective standardized clinical protocol. This enabled us to examine a large number of patients with early CCA with both imaging modalities. As previously described in a smaller study, most patients with PSC seen at Mayo Rochester outside the liver transplant practice had surveillance imaging (either ultrasound or MRI) at regular intervals.⁽⁶⁾ Imaging is performed on an annual basis among asymptomatic patients without cirrhosis. Given the differing study aims, participating centers, study intervals, and inclusion/exclusion criteria, only 12 subjects in the present study were included in this earlier study.⁽⁶⁾ Whether imaging was performed in symptomatic individuals, persons with PSC duration greater than 1 year, and those who had regular follow-up preceding their baseline ultrasound and MRI were also described.

DATA COLLECTION AND KEY DEFINITIONS

For this study, all CCA cases were perihilar and had either a characteristic mass lesion on cross-sectional imaging or adenocarcinoma on either brush cytology, biopsy, or explant tissue.⁽¹⁰⁾ The inclusion criteria for the perihilar CCA transplant protocol among those with PSC involves the following: individuals who would otherwise be a transplant candidate; radial diameter of mass 3 cm or less; no evidence of intrahepatic or extrahepatic metastases on cross-sectional imaging; or perihilar lymph node involvement on endoscopic ultrasound-guided fine needle aspiration performed before listing. The protocol involves neoadjuvant chemo-radiation and brachytherapy. While awaiting transplant, cross-sectional imaging is performed at 3-month intervals. Staging surgery is performed shortly before transplantation to rule out intrahepatic metastases or extrahepatic spread. After listing, individuals who develop intrahepatic or extrahepatic metastases (including lymph node involvement) are excluded from transplantation. After transplantation, there is a standardized protocol to monitor for CCA recurrence.⁽⁴⁾

The medical record was reviewed to determine the presence or absence of symptoms (abdominal pain, cholangitis, jaundice, pruritus) when imaging was obtained and at the time of CCA diagnosis. Patients who did not have symptoms suggestive of a biliary obstruction within the preceding 6 months were considered asymptomatic. The 6-month window prevented misclassification of individuals as being asymptomatic when they may have recently undergone successful endoscopic therapy for a flow-limiting stricture.

Laboratory data were collected at the time of the first imaging study. Carbohydrate antigen 19-9 (CA 19-9), aspartate aminotransferase (AST), and the serum alkaline phosphatase (SAP), which was divided by the upper limit of normal (ULN)–given variations in reference ranges, were also collected. Thresholds for SAP 1.5 times the ULN and CA 19-9 value greater than 100 U/mL were selected based on their prognostic relevance reported previously.^(4,11,12)

Importantly, radiologists who were blinded to all of the clinical information and original radiologic report re-examined the MRI (S.K.V. and C.L.W.) and ultrasound (S.P.S., I.S.I.) images for the purpose of this study and disagreements were resolved by discussion. The MRI protocol is described in Supporting Table S1, and all exams were done with gadolinium contrast and included MRCP. Images were classified as either (1) definite CCA, (2) possible CCA (should proceed with additional testing), or (3) negative for CCA (Fig. 1). MRI-only CCA detection was defined as an individual with CCA who had a MRI classified as either definite or possible for CCA and had a negative ultrasound. In contrast, ultrasound-detected CCA was defined as an individual with CCA who had an ultrasound classified as either definite or possible for CCA, regardless of the MRI result. A mass was defined as a well-circumscribed solid lesion. The radial diameter was categorized as either greater than 3 cm or less than or equal to 3 cm. Malignant strictures (Fig. 1) without a visible mass were classified as having a radial diameter less than or equal to 3 cm. This cutoff was selected because individuals with perihilar CCA and a mass with a radial diameter greater than 3 cm have a higher risk of adverse outcomes and are not eligible for MELD exception points.⁽⁴⁾

FIG. 1.

Representative images of CCA and radiographic criteria. (A-C) Definite CCA. Ultrasound images (A) showing hypoechoic mass (arrow) with peripheral duct dilatation. MRI postcontrast images in arterial (B) and delayed (C) phases showing progressive enhancement of the mass with maximum enhancement in delayed phase typical of CCA. (D-F) Possible CCA. Ultrasound images (D) showing common hepatic duct wall thickening and narrowing (arrow). MRI postcontrast images in arterial (E) and delayed (F) phases showing ductal wall thickening (arrow) with enhancement in both the arterial and delayed phases. Note that there is no distinct mass, soft tissue, or narrowing of vessels to suggest vascular encasement. Definitions: (1) MRI features of definite CCA (typical CCA features): (i) perihilar mass (with or without intrahepatic or intraductal mass lesions) with progressive enhancement on delayed phase imaging or (ii) malignant stricture (periductal focal soft tissue thickening with vascular narrowing or encasement or periductal soft tissue thickening that enhances on delayed phase imaging). (2) Ultrasound features of definite CCA: (i) periductal mass (with or without intrahepatic or intraductal mass lesions) or (ii) focal, nodular, shelf-like thickening of central duct with narrowing of the lumen with or without vascular narrowing or encasement. (3) MRI features of possible CCA (findings concerning for CCA; needs further evaluation): (i) stricture with nonfocal periductal thickening but without delayed phase enhancement or vascular narrowing (or encasement) or (ii) progressive lobar atrophy secondary to worsening perihilar stricture or (iii) ill-defined, irregular delayed enhancement of ductal wall without distinct mass. (4) Ultrasound features of possible CCA: (i) ductal dilation upstream from severe narrowing (parallel channel or shotgun sign) or (ii) lobar atrophy with dilated ducts and ductal crowding.

STATISTICAL ANALYSIS AND OUTCOME DEFINITIONS

Statistical analysis was performed with JMP and SAS software (SAS Institute; Cary, NC). All tests were two-sided with a level of significance of P < 0.05. Categorical data were compared using the Pearson chi-squared test, and continuous variables were compared using the nonparametric Wilcoxon test. Categorical data are presented as numbers (percentages), while continuous variables are expressed as medians, interquartile ranges. For the purposes of the statistical analyses, individuals whose imaging studies were categorized as either definite for CCA or possible CCA were considered positive for cancer unless otherwise stated.

We estimated receiver operating characteristic curves to determine the area under the curve (AUC) for each imaging category and its ability to detect CCA. In addition, the sensitivity, specificity, and positive and negative predictive values with respective 95% confidence intervals (CIs) were determined. The Fleiss kappa (κ) quantified the interrater agreement between the original radiologic diagnosis in the medical record and the blinded radiologic impression in the current study.

As previously described, dropout was defined as positive staging, CCA metastases, death, or withdrawal after listing but before transplant.⁽⁴⁾ Recurrence was defined as radiographic or histologic evidence of CCA following transplant.⁽⁴⁾ Subjects listed for transplantation were considered to have CCA progression if they met either the dropout or recurrence definitions.

The Kaplan-Meier method was used to describe survival free of events, and groups were compared with the log-rank test. Univariate and multivariate cox proportional hazard regressions were used to determine the association between covariates and endpoints. The outcomes of interest were death (all cause), CCA progression after listing, and CCA recurrence following transplant. For the former two endpoints, the baseline was at the time of CCA diagnosis, and censoring was at the last follow-up. For the latter endpoint, the baseline occurred at the time of transplant, and censoring at either death or last follow-up. Covariates included in the multivariable model were selected based on known associations with the endpoint of interest plus the absence or presence of symptoms at the time of CCA diagnosis.^(3,4,13,14)

Results

PATIENTS

Three hundred and fifty one patients with large-duct PSC who underwent both an ultrasound and MRI within 3 months of each other were identified. After excluding 85 individuals, 266 patients (asymptomatic, n = 127; symptomatic, n = 139) were included (Mayo Clinic Rochester, n = 248; Florida, n = 14; and Arizona, n = 4), and perihilar CCA was present in 120 individuals (Fig. 2). CCA and markers of cholestasis were more prevalent among those with symptoms (Table 1). Patients with CCA were more likely to be men, have a lower body mass index (BMI), shorter duration of PSC, and symptoms of biliary obstruction and elevation in liver biochemistries and CA 19-9 levels (Supporting Table S2). The cohort of persons without CCA was largely comprised and representative of persons with mild PSC, in whom 70% were asymptomatic with minimal elevations in liver biochemistries (Supporting Table S2). The duration of follow-up for the entire cohort was 6.79 (2.89-10.18) years.

FIG. 2.

Patients included. ^aAmong the 120 cancer cases, 107 were detected on imaging in the blinded radiology review.

TABLE 1.

Baseline Characteristics

	Entire Cohort (n = 266)	Asymptomatic Cohort (n = 127)	Symptomatic Cohort (n = 139)	P Value*
Age (years)	47.31 (35.27-56.09)	47.99 (34.85-57.73)	45.74 (35.31-55.56)	0.53
Female	32.33% (86 of 266)	33.86% (43 of 127)	30.94% (43 of 139)	0.61
BMI (kg/m2)	25.21 (22.49-28.34)	26.31 (23.59-29.89)	24.67 (21.94-27.46)	<0.01
PSC duration (years)	4.61 (0.96-9.70)	5.36 (2.12-9.86)	3.99 (0.22-9.67)	0.06
IBD present	78.95% (210 of 266)	76.38% (97 of 127)	81.29% (113 of 139)	0.33
CCA present	45.12% (120 of 266)	18.90% (24 of 127)	69.06% (96 of 139)	<0.0001
Symptoms†	52.25% (139 of 266)	—	100% (139 of 139)	—
Abdominal pain	22.55% (60 of 266)	—	43.17% (60 of 139)
Cholangitis	15.41% (41 of 266)	—	29.50% (41 of 139)
Pruritus	20.68% (55 of 266)	—	39.57% (55 of 139)
Jaundice	30.45% (81 of 266)	—	58.27% (81 of 139)
SAP/ULN	2.47 (1.45-4.13)	1.98 (1.03-3.18)	2.98 (2.03-5.12)	<0.0001
AST (U/L)	72.00 (40.00-113.00)	49.00 (33.50-85.50)	92.00 (56.00-146.00)	<0.0001
Total bilirubin (mg/dL)	1.20 (0.50-3.00)	0.80 (0.60-1.40)	2.50 (1.10-10.28)	<0.0001
Platelets (×109/L)	262.00 (202.00-338.50)	243.00 (194.00-304.50)	276.00 (203.00-376.00)	0.03
CA 19-9 (U/mL)	29.00 (12.00-105.00)	18.00 (9.00-35.00)	53.00 (19.50-275.25)	<0.0001
MELD score	9.96 (6.79-14.98)	7.17 (6.43-10.01)	11.32 (7.81-19.30)	<0.0001

^*Asymptomatic versus symptomatic.

^†Multiple symptoms may be present in the same patient.

Abbreviation: IBD, inflammatory bowel disease.

The cohort is representative of early-stage CCA (Table 2), and 93.33% (112 of 120) were listed for liver transplantation and received neoadjuvant chemotherapy and radiation. The presence of a low CA 19-9, low SAP, and normal bilirubin frequently occurred among patients with early-stage CCA, particularly among those who were asymptomatic (Table 2). Indeed, most of the patients with asymptomatic CCA had a normal bilirubin and other favorable prognostic features suggestive of an earlier stage of malignancy, including a lower CA 19-9 and a lower proportion of individuals with a discrete mass at the time of diagnosis or residual cancer on their explant (Table 2).

TABLE 2.

Features of Patients With CCA

	All Cancer Cases (n = 120)			Cancer Cases Detected on Imaging (n = 107)*
	Asymptomatic	Symptomatic	PValue	MRI Only	Ultrasound†	PValue
Tumor features
Discrete mass‡	41.18% (7 of 17)	70.00% (63 of 90)	0.02	54.55% (24 of 44)	73.02% (46 of 63)	0.05
Radial diameter > 3 cm‡	0% (0 of 17)	7.78% (7 of 90)	0.11	0% (0 of 44)	11.11% (7 of 63)	0.02
Baseline advanced malignancy§	0% (0 of 24)	2.08% (2 of 96)	0.34	2.27% (1 of 44)	0% (0 of 63)	0.18
Adenocarcinoma on cytology or biopsy	100.00% (24 of 24)	89.58% (86 of 96)	0.03	97.78% (43 of 44)	85.71% (54 of 63)	0.04
Adenocarcinoma on explant	36.84% (7 of 19)	62.50% (35 of 56)	0.05	61.29% (19 of 31)	55.88% (19 of 34)	0.66
Microvascular invasion on explant	0% (0 of 19)	7.14% (4 of 56)	0.12	3.23% (1 of 31)	8.82% (3 of 34)	0.35
CA 19-9 (U/mL)	26.00 (10.00-112.00)	108.50 (24.25-448.00)	<0.01	41.05 (19.50-149.50)	187.00 (26.00-554.00)	0.02
CA 19-9 ≤ 100 (U/mL)	70.83% (17 of 24)	47.92% (46 of 96)	0.04	68.18% (30 of 44)	38.10% (24 of 63)	<0.01
Clinical features
Age (years)	46.15 (36.80-56.27)	47.52 (38.77-56.68)	0.67	44.54 (34.90-57.27)	49.44 (40.28-57.20)	0.19
Female	16.67% (4 of 24)	26.04% (25 of 96)	0.32	20.45% (9 of 44)	26.98% (17 of 63)	0.44
BMI (kg/m2)	26.25 (22.78-30.84)	24.08 (21.73-27.07)	0.13	25.09 (21.75-28.17)	23.96 (21.63-26.60)	0.28
PSC duration (years)	2.43 (0.65-9.07)	3.49 (0.18-10.05)	0.88	4.08 (0.08-11.35)	2.13 (0.18-8.63)	0.66
IBD present	79.17% (19 of 24)	81.25% (78 of 96)	0.89	81.82% (36 of 44)	79.37% (50 of 63)	0.75
Listed for transplant	91.67% (22 of 24)	93.75% (90 of 96)	0.72	93.18% (41 of 44)	95.24% (60 of 63)	0.65
Laboratory features
SAP/ULN	1.89 (1.07-3.20)	3.09 (2.23-5.25)	<0.01	2.71 (1.58-5.02)	3.18 (2.27-5.72)	0.08
SAP < 1.5/ULN	41.67% (10 of 24)	11.58% (11 of 95)	<0.01	22.73% (10 of 44)	8.06% (5 of 62)	0.03
SAP normal	16.67% (4 of 24)	4.21% (4 of 95)	0.04	13.64% (6 of 44)	0% (0 of 62)	0.001
AST (U/L)	42.50 (33.50-77.75)	95.00 (59.75-146.00)	<0.001	75.00 (45.50-103.25)	98.00 (60.00-154.50)	0.02
Total bilirubin (mg/dL)	0.80 (0.50-0.93)	4.50 (1.20-14.20)	<0.001	1.65 (0.80-9.80)	3.00 (1.20-13.50)	0.21
Total bilirubin normal	83.33% (20 of 24)	19.79% (19 of 96)	<0.001	40.91% (18 of 44)	20.63% (13 of 63)	0.02
MELD score	6.89 (6.43-10.73)	13.50 (8.65-20.27)	<0.001	10.00 (6.61-19.73)	13.30 (7.86-20.73)	0.21

^*Definite or possible radiology categories considered positive. There were 13 cases missed on both ultrasound and MRI.

^†All cases detected on ultrasound were also detected on MRI.

^‡Measured on MRI. Only includes individuals with mass that was able to be measured on MRI from blinded radiologic review. Malignant strictures without discrete mass were classified as ≤3 cm.

^§Intrahepatic metastases or extrahepatic spread.

Abbreviation: IBD, inflammatory bowel disease.

DIAGNOSTIC PERFORMANCE AND INTERRATER VARIABILITY OF ULTRASOUND AND MRI

Compared with ultrasound, MRI had an improved diagnostic performance to detect early-stage perihilar CCA in the entire cohort, among asymptomatic individuals and in those who were eligible for the liver transplant protocol (Table 3). There were no subjects in whom CCA was visualized on ultrasound but missed on MRI. Combining the definite and possible CCA radiologic criteria resulted in an improved sensitivity with a slight reduction in specificity compared with definite criteria alone. This approach mirrors clinical practice in which indeterminate but concerning imaging features may prompt an ERCP with biliary brushings or close follow-up interval imaging. Incorporating a CA 19-9 level greater than 100 u/mL along with imaging findings increases the sensitivity of CCA detection (Supporting Table S3). However, this comes at the expense of lowering the specificity, and the use of CA 19-9 did not enable ultrasound to achieve diagnostic parity with MRI even among asymptomatic individuals.

TABLE 3.

Diagnostic Performance of Ultrasound and MRI for CCA Detection

	Sensitivity % (95% CI)	Specificity % (95% CI)	PPV % (95% CI)	NPV % (95% CI)	AUC	TP	TN	FP	FN
MRI (entire cohort)
Definite or possible	89.17 (82.19-94.15)	85.62 (78.86-90.87)	83.59 (77.34-88.38)	90.58 (85.14-94.16)	0.87	107	125	21	13
Definite	58.33 (48.98-67.26)	97.95 (94.11-99.57)	95.89 (88.29-98.63)	74.09 (69.80-77.97)	0.78	70	143	3	50
Ultrasound (entire cohort)
Definite or possible	52.50 (43.18-61.69)	89.73 (82.93-93.65)	80.77 (71.63-87.48)	69.68 (65.39-73.65)	0.70	63	131	15	57
Definite	16.67 (10.49-24.56)	99.32 (96.24-99.98)	95.24 (73.14-99.32)	59.18 (57.21-61.13)	0.58	20	145	1	100
MRI* (asymptomatic cohort)
Definite or possible	70.83 (49.91-87.38)	89.32 (81.69-94.55)	60.7 (45.53-74.08)	92.93 (87.53-96.09)	0.81	17	92	11	7
Definite	29.17 (12.62-51.09)	100.00 (96.48-100.00)	100.00 (-)	85.83 (82.24-88.68)	0.65	7	103	0	17
Ultrasound* (asymptomatic cohort)
Definite or possible	29.17 (12.62-49.39)	89.32 (81.69-94.55)	38.89 (21.60-59.51)	84.40 (80.59-87.59)	0.59	7	92	11	17
Definite	0.00 (0.00-14.25)	99.03 (94.76-99.98)	0 (-)	81.10 (80.81-81.39)	0.50	0	103	1	24
MRI† (eligible for transplant protocol)
Definite or possible	90.18 (83.11-94.99)	85.62 (78.86-90.87)	82.79 (76.32-87.77)	91.91 (86.59-95.24)	0.88	101	125	21	11
Definite	58.93 (49.24-68.14)	97.95 (94.11-99.57)	95.65 (87.66-98.55)	75.66 (71.32-79.53)	0.78	66	143	3	46
Ultrasound† (eligible for transplant protocol)
Definite or possible	53.57 (43.90-63.55)	89.73 (83.62-94.13)	80.00 (70.62-86.94)	71.58 (67.21-75.59)	0.71	60	131	15	52
Definite	16.96 (10.53-25.22)	99.32 (96.24-99.98)	95.00 (72.08-99.29)	60.92 (58.89-62.92)	0.58	19	145	1	93

^*Only included patients who did not have symptoms of a biliary obstruction within 6 months of imaging.

^†Patients with CCA who were ineligible for the transplant protocol were excluded (i.e., individuals with advanced malignancy).

Abbreviations: FN, false negative; FP, false positive; NPV, negative predictive value; PPV, positive predictive value; TN, true negative; TP, true positive.

Individuals who were harboring CCA that was not detected on either ultrasound or MRI (n = 13) were more likely to have features suggestive of a reduced tumor burden, including a lower CA 19-9 and lower SAP or bilirubin (Supporting Table S4). Nearly 50% (57 of 120) of cancers were missed on ultrasound (Table 3).

Among asymptomatic persons, false-positive results occurred in a similar proportion between ultrasound and MRI (Table 3). Of the 11 asymptomatic individuals with a false-positive MRI (definite criteria, n = 0; possible criteria, n = 11), 6 did not have an elevated CA 19-9 or worsening sub-clinical liver biochemistries that may have otherwise prompted an ERCP. If the original radiologic diagnosis was considered (rather than the results of the blinded review given in Table 3), 1.5% (4 of 266) of individuals underwent an unnecessary ERCP due to a false-positive MRI finding alone (i.e., did not have other features suggestive of progressive flow-limiting biliary stricture where an ERCP may have otherwise been indicated).

MRI interpretation for the entire cohort was associated with less interrater variability compared with ultrasound. For the definite category, the Fleiss κ for MRI was 0.73 (95% confidence interval [CI], 0.63-0.82) and 0.31 (95% CI, 0.11-0.50) for ultrasound. Similarly, the Fleiss κ for combined definite and possible categories was 0.87 (95% CI, 0.81-0.93) for MRI and 0.48 (95% CI, 0.36-0.60) for ultrasound. Indeed, compared with the original radiologic diagnosis among the entire cohort, the performance of ultrasound (definite or possible category) improved in the blinded review (AUC 0.70 vs. 0.63) but was similar for the ultrasound definite category (AUC 0.58 vs. 0.58), MRI definite or possible category (AUC 0.87 vs. 0.86), and MRI definite category (AUC 0.78 vs. 0.78).

PROGNOSTIC SIGNIFICANCE OF ASYMPTOMATIC CHOLANGIOCARCINOMA DETECTION

The summary of clinical endpoints following a diagnosis of CCA is found in Supporting Table S5. Patients with CCA diagnosed before the onset of symptoms had an improved 5-year overall survival (82% vs. 46%, log rank P < 0.01) (Fig. 3A). The survival benefit of being asymptomatic at the time of CCA detection persisted in the multivariable analysis (hazard ratio [HR], 0.29; 95% CI, 0.10-0.83; P = 0.02) (Supporting Table S6).

FIG. 3.

Symptoms and clinical outcomes among those with CCA. (A) Overall survival. (B) Progression-free survival after listing for transplant. (C) Recurrence-free survival after liver transplant. Unadjusted HRs are shown.

Persons diagnosed with biliary cancer while asymptomatic and listed for transplant were more likely to remain free from CCA progression 5 years after their cancer diagnosis (77% vs. 37%, log rank P < 0.01) (Fig. 3B). This benefit continued in the multivariable analysis (HR, 0.27; 95% CI, 0.10-0.78; P = 0.01) (Supporting Table S6). Similarly, individuals with asymptomatic CCA had an improved 5-year recurrence-free survival following liver transplant (89% vs. 65%, log rank P = 0.04) (Fig. 3C). A trend toward asymptomatic persons having an improved recurrence-free survival following liver transplant was noted in the adjusted analysis (HR, 0.27; 95% CI, 0.06-1.16; P = 0.07) (Supporting Table S6).

PROGNOSTIC SIGNIFICANCE OF CCA DETECTION WITH MRI

Among the 120 cancer cases, 107 were detected on imaging. Compared with ultrasound-detected CCA, individuals with CCA detected only on MRI (i.e., missed on ultrasound) were more likely to have features suggestive of an earlier cancer, including a smaller tumor diameter, a lower CA 19-9, and reduced markers of cholestasis (Table 2).

Patients with asymptomatic CCA detected only on MRI had an improved 5-year survival compared to those with asymptomatic CCA detected on ultrasound (100% vs. 33%, log rank P = 0.05) (Fig. 4A). Similarly, individuals with asymptomatic CCA detected only on MRI were more likely to remain free from CCA progression after 5 years compared to those with asymptomatic CCA detected on ultrasound (88% vs. 33%, log rank P = 0.03) (Fig. 4B). However, the prognostic benefit of MRI detection over ultrasound was nullified after the onset of symptoms (Fig. 4A,B). MRI-only detection in asymptomatic persons did not translate into a reduction in CCA recurrence following liver transplantation (Fig. 4C).

FIG. 4.

Imaging modality and clinical outcomes among those with CCA detected on blinded image review. (A) Overall survival. (B) Progression-free survival after listing for transplant. (C) Recurrence-free survival after liver transplant. Unadjusted HRs are shown.

SENSITIVITY ANALYSES: PSC DURATION BEYOND 1 YEAR AND THOSE WITH ANNUAL IMAGING PRECEDING CANCER DIAGNOSIS

To further examine the prognostic benefit of CCA detection before the onset of symptoms and using MRI, we examined two patient subgroups. First, only individuals who developed biliary cancer 1 or more years after their PSC diagnosis were examined (n = 73). Indeed, CCA detection before the onset of symptoms continued to be associated with improved outcomes (Supporting Fig. S1). Moreover, there was a continued trend toward improved survival and a reduction in CCA progression following transplant listing among those asymptomatic individuals with MRI-only detected CCA (Supporting Fig. S2).

Second, we examined individuals who developed perihilar CCA at least 1 year after their PSC diagnosis who also had an annual follow-up preceding their cancer detection and either underwent yearly CCA imaging with any modality (n = 29) or did not undergo annual imaging surveillance (n = 6). Persons who participated in an annual imaging program (asymptomatic and symptomatic) before their cancer diagnosis had a reduction in mortality, CCA progression after listing, and CCA recurrence following liver transplant (Fig. 5). Asymptomatic persons derived the most benefit from this approach. For example, individuals who were asymptomatic with regular surveillance had an improved survival and a trend toward reduced CCA progression compared with their counter parts with symptoms who either did or did not participate in annual imaging preceding their cancer diagnosis (Supporting Fig. S3). Moreover, asymptomatic subjects in this subgroup continued to have improved outcomes when their CCA was detected only on MRI (compared to ultrasound) with a reduction in CCA progression after transplant listing (HR, 0.21; 95% CI, 0.05-0.99; P = 0.04) and a trend toward reduced mortality (HR, 0.13; 95% CI, 0.01-1.28; P = 0.08).

FIG. 5.

Impact of annual imaging on outcomes among those with CCA detected 1 year after PSC diagnosis. (A) Overall survival. (B) Progression-free survival after listing for transplant. (C) Recurrence-free survival after liver transplant. Unadjusted HRs are shown.

Discussion

Before this study, there was limited evidence regarding the diagnostic performance of ultrasound and MRI to detect early-stage perihilar CCA in PSC, uncertainty if it was advantageous to diagnose CCA before the onset of symptoms, and no known prognostic benefit of using MRI compared with ultrasound. However, the present study sheds light on these issues and has several observations with direct clinical implications. First, MRI (rather than ultrasound) is better suited to discover early-stage CCA, particularly among patients who are asymptomatic and eligible for curative therapy. Second, individuals with CCA who are either asymptomatic or their cancer is visualized only on MRI (i.e., missed on ultrasound) are more likely to have favorable prognostic features. Third, patients who have their CCA detected before symptoms develop have improved outcomes compared to those in whom detection is delayed until the onset of symptoms. Finally, asymptomatic patients with CCA whose cancer was detected on MRI but missed on ultrasound had better outcomes.

MRI demonstrated an overall sensitivity of 89% and is the imaging method of choice to detect early-stage perihilar CCA in PSC. In asymptomatic persons, the sensitivity of MRI decreases to 71%. For perspective, the reported pooled sensitivity for ultrasound detection of HCC in those with cirrhosis was 84% for any stage and 47% for early-stage cancer.⁽¹⁵⁾ After completion of the study, the MRI radiologists (C.L.W. and S.K.V.) were unblinded and reviewed the false-negative cases among the asymptomatic cohort with CCA. No potential or definitive features of CCA were seen in the unblinded review, and it is noteworthy that the original radiologic report was also consistent with this impression (i.e., 3 radiologists agreed). This illustrates the challenge of early perihilar CCA detection. A potential harm of any test is a false-positive result. Indeed, marked periductal inflammatory thickening can lead to an indeterminate MRI result and was the reason for all false-positive MRIs among the asymptomatic group. In our clinical practice, we typically proceed with an ERCP and obtain biliary brushings for routine cytology and FISH when an indeterminate biliary stricture is encountered on MRI. Previously, it has been suggested that ultrasound can be used as a first-line test rather than MRI for CCA surveillance.^(8,9) However, our results indicate that this approach would fail to detect a significant number of early-stage malignancies and miss a finite window of opportunity to implement a curative treatment plan. Indeed, when CCA was able to be visualized on ultrasound, the tumor was already larger and patients had features suggestive of a more advanced cancer (Table 2). This is critical, as a discrete mass with a radial diameter greater than 3 cm is associated with poor outcomes and precludes patients from proceeding with liver transplantation.⁽⁴⁾ Hence, detection while the radial diameter remains small is important. Moreover, the sensitivity for ultrasound to detect CCA in asymptomatic persons was only 29%. For comparison, in a smaller cohort that had a higher prevalence of advanced cancer, the sensitivity of ultrasound was 57%.⁽⁵⁾ Thus, while ultrasound has the capacity to detect CCA, it lacks sufficient sensitivity to detect disease when curative action can be taken.

Clinicians should consider the routine use of MRI in patients with PSC to detect CCA, regardless of symptoms or laboratory tests suggestive of a progressive biliary obstruction. For example, asymptomatic CCA is not uncommon, and detection of cancer in this subgroup is feasible and associated with improved survival and both progression-free and recurrence-free survival. Indeed, individuals diagnosed with CCA while asymptomatic had a 36% reduction in 5-year mortality when compared with their symptomatic counterparts. In turn, CCA detection before the onset of symptoms was independently associated with improved outcomes after adjusting for other relevant prognostic variables (Supporting Table S6). Moreover, laboratory tests alone are inadequate to detect early cancer as having a low bilirubin, SAP, or CA 19-9, while harboring an occult CCA is not uncommon (Table 2). In addition, the enhanced diagnostic performance of MRI (Table 3) translates into improved outcomes (Fig. 4). Indeed, MRI detection (compared with ultrasound) was associated with a 77% reduction in 5-year mortality among asymptomatic persons. However, the benefit of MRI detection over ultrasound becomes attenuated after the onset of symptoms. Likely this is because MRI was better suited to detect CCA earlier, and its impact on outcomes was mitigated as the tumor progressed and caused symptoms. Importantly, these prognostic observations persisted after excluding individuals in whom CCA was detected within 1 year of their PSC diagnosis. Moreover, despite having less than 5% patient overlap, we reaffirmed an earlier observation from a smaller study in our center that regular imaging is beneficial,⁽⁶⁾ but further noted that asymptomatic individuals undergoing surveillance and having their cancer detected on MRI alone incurred the most benefit. Taken together, these observations suggest using MRI surveillance (rather than ultrasound) to test for occult CCA before the onset of symptoms is both feasible and can translate into improved outcomes.

This study has several strengths. First, it represents a large cohort of patients with PSC diagnosed with perihilar CCA at an early stage. This affords a unique opportunity to systematically study a complication of a rare disease. Moreover, it delineates asymptomatic persons from those with symptoms. This is important, as performing imaging in individuals with a new symptomatic biliary obstruction is a diagnostic study rather than a routine surveillance test. Second, it studies patients with early CCA who underwent both ultrasound and MRI within the same time period, which enables the direct comparison of these imaging modalities. Third, imaging was obtained through a prospective clinical protocol, and CCA treatment and follow-up testing for cancer recurrence and adverse outcomes were standardized. This allowed us to examine the true impact of early detection and outcomes without additional confounders such as variations in testing, treatment, and monitoring protocols. Fourth, it is important to note that all images were re-examined by radiologists blinded to all clinical data and radiologic reports to minimize bias. In contrast, previous PSC-CCA studies examining ultrasound and MRI used radiologic data abstracted retrospectively from imaging reports that were created by radiologists who were aware of all the patients’ clinical information (symptoms, clinicians’ impressions, and whether there was a concern for CCA).^(5,6)

This study has several limitations. First, although a multicenter randomized controlled trial would be the ideal study design to minimize bias, this would be challenging to conduct for a variety of reasons. While the study is enriched with patients with early CCA, the present study has several attributes that mitigate the potential impact of bias, including selection bias, on our conclusions: (1) In contrast to the positive and negative predictive values, the sensitivity and specificity of ultrasound and MRI may be less influenced by the prevalence of CCA in the study population; (2) we assessed cancer-related outcomes, and the pool of patients without CCA would not influence these results; (3) the cancer cohort was representative of early-stage perihilar CCA and reflects the many patients with PSC CCA who underwent a standardized treatment protocol in the United States; (4) both ultrasound and MRI were obtained concomitantly as part of a prospective standardized clinical protocol for patients suspected of early CCA; (5) radiologists were unaware of the patients’ information, including whether there was a pre-existing concern for CCA; and (6) our results were confirmed in subgroup analyses, including in those with annual imaging preceding cancer detection. Second, this study was conducted at academic centers, and some of our results may not be generalizable to institutions that infrequently see patients with PSC, use an alternative treatment for early-stage CCA with differing efficacy (including areas with limited access to liver transplantation), or where there are greater variations in imaging interpretation or quality. However, most patients with PSC and CCA are ultimately seen at academic centers, and it is not feasible to conduct such a study without drawing from those referral practices. This approach enabled us to maintain quality consistency with in-depth data collection and patient-care practices. Furthermore, our observations that an earlier stage of disease was present in persons who were asymptomatic and had their CCA detected on MRI are independent of our treatment approach and have clinical implications for areas where access to transplantation is limited. In addition, it is known that both imaging modalities can be subject to variations in quality and interpretation. To minimize this, an international consortium of radiologists and hepatologists published recommended quality standards when using MRI in those with PSC.⁽²⁾ Our results suggest that MRI when reviewed by radiologists is subject to less variation in interpretation compared with ultrasound, which, coupled with its superior diagnostic performance, further reinforces its role in early CCA testing. Third, the diagnosis of CCA is challenging, and not all patients may have adenocarcinoma or definitive imaging features detected. Therefore, some cases of subtle early CCA may not have been included in the present study. However, when we examined individuals who did not meet our definition of CCA yet had an indeterminate stricture with serial or multifocal polysomy plus another additive factor (elevated CA 19-9 or cytology suspicious for adenocarcinoma) suggestive of CCA, the principle results were unchanged (data not shown).

Our results highlight that future studies examining early CCA detection and surveillance should delineate between asymptomatic and symptomatic persons, and make the distinction between diagnostic testing in patients with new symptoms and routine asymptomatic surveillance. Key unanswered questions and areas for further study include defining the optimal surveillance interval and cost effectiveness of using MRI for early CCA detection for a general PSC population, refining our ability to risk-stratify patients, and developing noninvasive tests that are better suited to diagnose CCA. The discovery of new cancer biomarkers and the use of advanced imaging analytics, such as deep learning, may hold promise but require investigation and validation in large cohorts with PSC and early-stage perihilar CCA.^(16,17)

In summary, we present evidence that indicates MRI should be used as the imaging method of choice to detect early perihilar CCA in those with PSC, and providers should strive to diagnose biliary cancer before the onset of symptoms. Therefore, routine testing for CCA among persons with PSC, regardless of symptoms, should be considered. In the absence of an effective medical therapy for PSC, early detection and treatment of the leading source of mortality creates an opportunity to have a positive impact on the natural history of this disease.

Abbreviations:

AST

aspartate aminotransferase

AUC

area under the curve

BMI

body mass index

CA 19-9

carbohydrate antigen 19-9

CCA

cholangiocarcinoma

CI

confidence interval

CT

computed tomography

ERCP

endoscopic retrograde cholangiopancreatography

FISH

fluorescence in situ hybridization

HCC

hepatocellular carcinoma

HR

hazard ratio

MELD

Model for End-Stage Liver Disease

MRCP

magnetic resonance cholangiopancreatography

MRI

magnetic resonance imaging

PSC

primary sclerosing cholangitis

SAP

serum alkaline phosphatase

ULN

upper limit of normal